The short version
- Contact discovery runs only after qualification, and only for domains that produced a qualified opportunity — never on raw search results.
- Work is deduplicated by publisher domain and each source is reused for 30 days, so five opportunities on one site cost one lookup.
- Public pages are checked first: the already-scraped article, then the homepage, /contact, /contact-us, /support, /help, /about, and /press, plus one explicitly linked editorial or team page. At most eight pages per domain.
- Hunter is used where it adds something — the identified author first, domain search only as a fallback — rather than on every candidate.
- Addresses are verified before a first send, because bouncing off dead addresses damages your own domain's sending reputation permanently.
Every email-finding provider charges per lookup. Run one for every URL a search returns and the bill arrives long before the links do — and most of those lookups were for pages you were never going to pitch.
The fix is not a cheaper provider. It is doing the lookup last, on the shortest possible list, from the cheapest source that works.
Ordering is the cost control
In Dice, contact discovery is invoked only after qualification has persisted qualified opportunities. A page that failed the hard checks, scored below the floor, or already links to you never reaches the contact step at all.
Then the list shrinks again, because contacts belong to publishers rather than pages. Work is deduplicated by publisher domain: five qualified opportunities on the same site is one contact discovery, not five. Each source is reused for 30 days, so the next run against that publisher costs nothing.
Contact discovery order
Runs only after an opportunity qualifies · deduplicated per publisher domain
01The article, already fetched
No extra costAuthor byline and any address published on the page itself.
02Public publisher pages
Fetch onlyHome, /contact, /contact-us, /support, /help, /about, /press, plus one explicitly linked editorial page. Eight pages maximum.
03Provider lookup: the identified author
CreditsA targeted search for the person who wrote the page.
04Provider lookup: domain search
CreditsFallback only, when no author is identified or found.
Verification before the first send. Provider-supplied addresses are verified before outreach goes out. Addresses published on a publisher’s own contact page are not pushed through a verifier by default — several credits per publisher for very little information.
Public pages first
A large share of publishers publish an address in plain sight. Checking for it costs a page fetch and no provider credits, so it happens first.
The article itself is reused rather than re-fetched — enrichment already paid for it — and then a bounded set of standard locations is checked:
- The homepage
/contactand/contact-us/supportand/help/about/press- One editorial, team, or contributors page, but only when the site explicitly links to it
That last rule is the one that keeps this bounded. Following explicit links only, rather than guessing at paths, means the scan never turns into an unbounded crawl of someone else's site. The whole pass is capped at eight pages per domain, and up to eight distinct addresses are retained, each stored with the page it was found on.
What gets filtered out
Scraping text for anything shaped like an email address produces a lot of things that are not usable contacts. Several classes are discarded or demoted:
| Class | Treatment |
|---|---|
| Unattended addresses | no-reply, do-not-reply, mailer-daemon, and postmaster are dropped entirely. Nobody reads them. |
| Documentation examples | Addresses at example.com and its siblings are dropped — they are placeholders in someone's code sample. |
| Generic role addresses | hello, info, support, editorial, press and similar are kept but marked generic, because they are real and less valuable than a named person. |
| Off-domain addresses | Addresses that do not match the publisher's domain are kept with that fact recorded, since they are weaker evidence of who runs the site. |
| mailto links | Treated as stronger evidence than a plain text match — someone deliberately published that address as a way to be contacted. |
Each surviving address carries a confidence value, whether it is generic, whether it matched the publisher's domain, and the page it came from. That detail is what reachability scoring consumes.
Hunter, used surgically
When public pages do not produce a usable contact, Hunter fills the gap — but the order of operations matters as much as the provider does.
- 01
The identified author first
Enrichment often identifies who wrote the page. Looking up that specific person by name at that domain is a targeted query with a high hit rate and a clear answer.
- 02
Domain search only as a fallback
When there is no author, or the author cannot be found, a domain search returns whoever is known at the publisher. Broader, less precise, and used second for exactly that reason.
- 03
Verification before the first send
Provider-supplied addresses are verified before outreach goes out. This is not optional: sending to dead addresses is how a sending domain acquires a reputation problem it does not recover from.
Bouncing off dead addresses destroys sending reputation, and reputation is the one asset in cold outreach you cannot buy back.
One provider is enough to start
Public page scanning needs a scraping provider; Hunter lookups need a Hunter key. Either alone produces contacts — the public scan finds editorial and support inboxes, Hunter finds named people. Configuring both simply means fewer opportunities end up with nobody attached.
Opportunities that end up with no contact at all are not discarded. They rank lower, they keep their full evidence, and they can pick up contacts on a later pass — because the ranking reruns globally whenever contacts change.
Once you have contacts, the next question is what to say. That is outreach drafts grounded in the article.